I saw this a few days ago and ignored it at first. Vendor bashing pieces are nothing new from the media. To paraphrase Mark Twain: Rumors of Dell’s demise are greatly exaggerated. Dell is, and continues to be a powerhouse at pushing machines out. Their innovation is, basically customized machines in volume. There is no magic in their machines, they use the same parts from the same sources as others do.

This is going out to a customer in about a week, though we have a little time to run tests. Tuning these units is like tuning an engine … and you like to tweak and tune, you eventually learn where the parameters of this engine, from tuning, hit good efficiency. You get a feel for it. You tune one parameter, the vibration pattern changes slightly. You tune another parameter, you get another overtone.

This discussion erupted on the beowulf list today. I responded to a question on this, pointing out that prestige adds nothing to the bottom line. What matters is, not so curiously, the bottom line. One author disagreed with me. His point was that prestige class systems translated into sales for the relevant vendors. I think his examples were stretches, and not applicable to HPC. All of the public (and private) responses I have seen seem to support my thesis on this.

that basic systems management and configuration are neither well known nor well understood, by the vast majority of people. Even the ones with certifications who should know this stuff. This is leading me to rethink some of the basic elements of the out-of-box experience. Part of this is driven by the tendency of certain distributions to remap, effectively randomly, their network port assignments. This is never, under any circumstances, a good thing.

Running some burn-in testing as I remember (older video) … [FLOWPLAYER=/wp-content/videos/siCluster_in_action.flv,320,200] You can just feel the bits-a-flowing … paraphrasing Apocalypse Now “I love the smell of many GB/s in the morning …”

/. linked to a newly granted patent by Amazon. I am all for good patents, but given the number of … er … not so good ones I’ve read through, as well as obvious rehashing of existing work … I don’t know what to make of this one. This seems … well … like patenting any instance of a market exchanging money for computer time and/or storage, based upon a pricing model determined by past histories or current demand/availability.

A customer with a RAID6 and RAID1 OS drive just had what amounts to an epic failure. 4 drives gone. These are pretty good drives, not a known bad batch. Data points to environmental issues (heat). They don’t understand it, given the nice AC in there, but looking at the drives, a number of them were warm. We have fans doing a pull across the drives, they shouldn’t have been as warm as they were.

Strong storms ran through the area Friday night. And it messed with TV and internet. The former, not so concerned about. The latter … scalability.org runs from a home machine. We don’t generate revenue from it, and I am not willing to host it on a provider so I don’t want it to be marginally more costly than our internet service plus our home power to run the server (relatively light BTW).

Oracle setting Solaris on HP boxen. Oracle wants the hardware revenue too … though it could also be a step to dropping x64 platforms for Solaris. I can’t see how this is a good move for Solaris ubiquity, given the ability of HP to move hardware. And the name of the game for growing OS support is … curiously … ubiquity.

This is annoying. Intel’s compilers do unbuffered IO by default (as in out of the box). Which means if you have code like this: #define BIG 1000000000 real*8 X(BIG) do i=1,BIG write (unit=10,*) X(i) enddo then you are going to suffer terrible performance, as Fortran (Intel’s compiled version) will do a flush at the end of each write. Which means, for a high performance network file system, you are going be hitting it with many ~25 byte writes.